Convolutional Neural Network Based Antenna Beam Coefficient Generation for Planar Arrays

Abstract

This work presents a novel design approach on optimizing the complex feeding coefficients of an antenna array to achieve a desired beam pattern. The approach uses a convolutional neural network which takes an image representation of a desired radiation pattern and generates the corresponding phase gradient over the array aperture. To demonstrate the performance of the approach, an 8×8 planar array has been used. The results show the potential for machine learning to optimize antenna parameters in more complicated antenna systems, where an efficient way of developing antenna pattern codebooks by using a trained neural network is used.